Your best bet is to use C. It is highly efficient. If possible use computational code like the Atlas BLAS package. This code will run circles around your own code no matter what language you use. You already know C and moving to C++ is a major problem. All the other languages are distractions from your purpose.
If possible run multiple, independent processes rather than writing parallel code. That can be a major ordeal.
If your goal is to process data as opposed to learning elaborate programming techniques, keep simplicity in mind. C is a very powerful language and you can reach maximal efficiency for many problems using Atlas BLAS and multiple processes. If you goal is to get a degree in CS, ignore what I've suggested.
I think $54M is a bit expensive too. I can imagine that it might take some effort, but this would pay a team of 500 programmers for a year. I expect it could be done by a team of 5 to 10 in the same time. This sounds like the mythical $500 hammer to me.
A better alternative would be to set up no exchanges and let the insurance companies provide the web sites. You would find some innovative people setting up exchanges on their own just like you now find exchanges for hotels and airlines. All the government needed to do was tell the insurance companies the new insurance laws and let it happen.
If you're having performance issues, then C++ would offer a more efficient solution. Why jump through all these hoops to boost Java performance? Just use C++ and get twice the performance instantly with Linux. I tend to agree with the AC that the language issue is overblown. With practice programming is about the same level of difficulty with most languages. C++ does a pretty good job at compile time checking, interfaces directly with the system calls and offers nearly all the performance you can get from the computer. (AVX instructions can be done quite well in assembly, but most cloud apps would not benefit from using AVX.)
Unless you have substantial computation or disk I/O, I would expect the bottleneck to be the network. With compute bound apps the OS is irrelevant. Likewise with I/O devices the time spent in system calls is dwarfed by the disk speed.